What is AgentKanban and why is it being talked about

AgentKanban is a tool for managing context and workflow in projects where developers collaborate with AI agents. Instead of Copilot conversations remaining confined to a single chat window, AgentKanban automatically saves individual "turns" — meaning user prompts and agent responses — directly to the relevant task on a shared Kanban board. This gives the team a clear history of decisions that survive even after closing the editor.

The tool is based on the practice of the British company App Software Ltd and currently offers a web application along with an extension for Visual Studio Code. It also supports VS Code-compatible editors such as Cursor or Windsurf. Developers can thus keep their favorite harness — GitHub Copilot is currently fully integrated, with other agents under development.

Planning, TODOs, and Implementation on One Board

AgentKanban emphasizes a structured approach to working with AI. The workflow consists of three phases: planning, breaking down the task into TODO items, and implementation. In the first phase, the team jointly creates and refines the assignment on a real-time Kanban board. The agent then has space to ask clarifying questions before starting work.

In the second phase, the AI breaks down the task into iterations and individual TODO items, which can be tracked directly on the board. Both the developer and the agent can see what is already done and what remains. During implementation, the agent follows the approved plan and continuously updates the status of individual items. All comments, attachments, and technical notes remain with the task, not scattered across chat windows.

Integration with VS Code and GitHub Copilot

A key advantage is the VS Code extension, which connects the editor with the hosted board. After selecting an active task, the extension takes care of inserting context into the GitHub Copilot chat using the command @kanban /resume. This loads the board's memory, task description, event timeline, and the current TODO list. If you only want to get new changes since the last resume, just use the variant @kanban /resume delta.

The extension automatically captures conversational turns and sends them to the server. Users can later view, pin, split, or delete individual turns. This context is then available to all team members who have access to the board. Git worktrees allow isolating each task into its own branch and separate directory, preventing unnecessary code conflicts.

The tool supports the Model Context Protocol (MCP) — an open standard by Anthropic for connecting AI agents with external data sources. Thanks to the automatically generated .vscode/mcp.json configuration, the agent can read and modify tasks, comments, TODO items, or board memory directly via API. This means that AgentKanban is not just a passive notepad, but an active part of the agent's workflow.

Pricing and Limits: From Solo Developer to Large Team

AgentKanban offers two subscription tiers. The Free plan is available for free without time limits and includes up to 3 team members, 3 boards, and 30 active tasks. The daily limit for API and MCP requests is 250 calls, attachment storage is 250 MB, and data is retained for 14 days. This is more than enough for personal projects, small teams, or testing the concept.

The Pro plan costs 10 USD (approximately 230 CZK) per user per month. An organization gets up to 100 boards, 10,000 tasks, 25 webhooks, and 50 API keys. The daily limit increases to 25,000 requests, storage to 50 GB, and data retention to 365 days. Attachments can be up to 100 MB in size. Billing is at the organizational level — the owner purchases seats and assigns them to members.

Security and Availability for Czech Developers

AgentKanban is available globally, including the Czech Republic. Both the web interface and the VS Code extension are in English; Czech localization is not yet available. However, this does not prevent Czech development teams from full utilization — the technical documentation is understandable, and the VS Code interface remains in English for other tools as well.

In terms of security, the tool supports TOTP two-factor authentication, backup codes, and modern passkeys via WebAuthn, Touch ID, or Windows Hello. Webhooks are signed using HMAC-SHA256, which ensures data integrity when integrating with external systems. The project thus meets standard security requirements for corporate workflows in the European environment.

Also worth mentioning is AGENTS.md — a concept that AgentKanban lends to the entire board. In the board settings, instructions for the AI agent can be defined, which are then automatically inserted into a local AGENTS.md file in the workspace. This means that the entire team shares the same context and conventions without having to manually copy instructions between projects.